shubhamdikshit

Configured and deployed shipment, weight discrepancy, COD reconciliation, and PO reconciliation modules. Developed and maintained Redis-based caching and MariaDB, optimizing data processing. Built scalable backend services using Rust and Golang, ensuring high performance. Worked with Kafka, and MongoDB for real-time data ingestion and processing. Assisted in UAT testing, ensuring seamless deployment and resolving defects based on client feedback.

Predicted credit card fraud using XGBoost (F1-score: 80%) and built churn prediction models using PCA & logistic regression. Used PCA along with logistic regression to build customer churn prediction model using scikit-learn package which gave a sensitivity of 75% in highly imbalanced data set. Automated 40% of the SOC answering process using BERT for text similarity. Created robust flask APIs and hosted it over Azure VM using Nginx and uWSGI. Created Front-end and backend components using React, JavaScript, and Python. Created anomaly detection rules for lead validation using PySpark and Azure Data Lake. Data analysis of all types of transactions from an individual or entity using SQL and python to identify anomalies and suspicious targets.

Conducted data analysis on survey responses to support R&D teams. Created monthly reports on sales trends and consumer behavior, providing insights for management. Assisted in data-driven marketing strategies, leading to increased sales and brand awareness.

Constructed an Explainable Artificial Intelligence (XAI) model for sarcasm detection in dialogues, achieving an accuracy of 95% on the MUStARD Dataset. Developed a pipeline employing a diverse range of algorithms, including Ensemble models, kernel-based models, Decision Trees, and Regression models, to assess their accuracy in decision-making. Applied the XGBoost machine learning algorithm to analyze a dataset comprising 690 dialogues from the MUStARD Dataset, achieving an accuracy of 90%. Applied LIME and SHAP explainability techniques to enhance the performance of the XGBoost algorithm, resulting in a notable 5% increase in accuracy to 95%.

Executed Language Translation of Low Resourced Indian Languages using Seq2Seq model using pytorch. Improved BLEU scores by 10% for Tamil (5.43), Telugu (4.97) and Bengali (4.07) languages. Used Attention Decoder to obtain features. Trained and deployed the model using PyTorch.

Computed translation probabilities between words and phrases of different Indian Languages, achieving an accuracy of 95%. Implemented Zipf's law in Hindi, Tamil, and Telugu corpora, reducing the size of the corpora by 50% without sacrificing accuracy. Built a custom parser for Hindi, Telugu, and Tamil, which is able to parse sentences with an accuracy of 90%.